A method for feature point recognition and positioning of PPG and its derived signals

Abstract

The present application relates to a kind of PPG and the feature point recognition and positioning of its derived signal method, belong to biomedical signal processing technical field.The method includes: S1: signal pre-processing: filtering and complete cycle extraction;S2: basic reference point detection: the method of moving sliding window and threshold combination is used to identify wave peak, wave trough, zero crossing point and other feature points mapped;S3: waveform classification: according to the basic reference point of detection as reference, based on the characteristics of waveform morphological change realizes classification;S4: different category feature point identification: according to the waveform classification result corresponding method is selected to realize the feature point identification in different category state.The present application realizes the identification and positioning of multiple categories, multiple feature points, can accurately identify and locate the feature point of PPG and its derived signal under pathological state.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of biomedical signal processing, and relates to a method for feature point recognition and positioning of PPG and its derived signals. BACKGROUND

[0002] The pulse is the external reflection of important information such as the state of the heart and blood vessels, and its fluctuation contains rich physiological and pathological information. Each beat of the cardiovascular system drives the periodic fluctuation of blood volume in peripheral blood vessels, and this dynamic change is the fundamental mechanism of PPG waveform formation. Any subtle change in the human body can have a significant impact on the pulse system, and the resulting changes are not only reflected in the amplitude, rate and rhythm of the waveform, but also reflect the current physiological state of the cardiovascular system and are sensitive indicators of potential pathological changes. Therefore, PPG waveform formation is a key window for exploring the physiological and pathological conditions of the human body.

[0003] At present, the feature point recognition method for pulse wave signals includes the time domain difference threshold method, the time domain discrimination method based on extreme value, amplitude and slope, and the method based on Hilbert transform, EMD and wavelet transform to decompose the signal and detect the characteristic waveform at a specific layer.

[0004] However, due to the complexity of pathological factors and the variability of signals in clinical practice, when faced with pathological pulse wave signals, the interference of noise and complex pathological state can easily cause the signal features to be submerged or the reference points to be blurred due to severe signal deformation, thereby making the traditional feature point recognition method limited, resulting in difficulties in feature point recognition, low recognition efficiency, poor recognition accuracy and other problems. SUMMARY

[0005] Therefore, the present application aims to solve the problems of difficult feature point recognition, low recognition efficiency and poor recognition accuracy of pathological pulse wave signals, and provides a method for feature point recognition and positioning of PPG and its derived signals based on multiple features and multiple categories, which is used for accurately recognizing and positioning the feature points of PPG and its derived signals in pathological state.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0007] A method for feature point recognition and positioning of PPG and its derived signals, specifically comprising the following steps:

[0008] S1: signal preprocessing: obtaining the original PPG signal, removing high-frequency noise and baseline drift using a Butterworth low-pass filter and a cubic spline interpolation method, and extracting the waveform in the time domain whole cycle range using a cycle segmentation method;

[0009] S2: Detection of basic reference points: Based on the time-domain whole-period PPG signal obtained in step S1, the waveform data of the PPG signal is subjected to point-by-point difference operation, the first derivative function VPG signal and the second derivative function APG signal of the PPG signal are calculated respectively, and the basic reference points of the PPG, VPG and APG waveforms are detected by using the method of moving sliding window, threshold judgment and adaptive dynamic adjustment strategy;

[0010] S3: Waveform classification: Based on the basic reference points identified in step S2, the signal segments are divided as reference, and the effective classification of two different forms of PPG waveforms and three different forms of APG waveforms is realized according to the classification strategy of waveform morphological changes;

[0011] S4: Recognition of feature points of different categories: Based on the different categories of PPG waveforms and APG waveforms obtained by the classification strategy of waveform morphological changes in step S3, the feature points matched with the waveform characteristics are adaptively selected to identify the N points and D points of different PPG waveforms, and the c points and d points of different APG waveforms; wherein the N points and D points are the two-point tangent points and diastolic peak points of the PPG waveform respectively; the c point is the late systolic re-increase wave of the APG waveform, and the d point is the late systolic re-decay wave of the APG waveform.

[0012] Further, in step S2, the basic reference points include the peak S and the trough O of the PPG waveform, the peak u and the trough v of the VPG waveform, the peak a and the trough b of the APG waveform, and the feature points of the peak and the trough of each waveform mapped to other waveforms, i.e. the maximum slope point E of the rising branch and the minimum slope point F of the falling branch of the PPG waveform, the zero-crossing point L of the VPG waveform, the diastolic peak w point of the VPG waveform, and the late systolic re-increase wave e point of the APG waveform.

[0013] Further, in step S2, the width value W of the moving sliding window is set according to the following formula:

[0014] W = m * fs

[0015] Wherein, fs is the signal sampling frequency, and m is a coefficient, whose value range is 0.6-1.2; the preset amplitude threshold H is 0.6-0.8 times of the maximum amplitude in the waveform data, and the peak and the trough of the PPG, VPG and APG waveforms are identified.

[0016] The adaptive dynamic adjustment strategy includes threshold judgment of peak interval and judgment mechanism for low-amplitude missed detection.

[0017] Further, in step S2, the threshold judgment of the peak interval is a method of combining a moving sliding window with an initial threshold to identify the peaks and troughs of the PPG, VPG and APG signals, and then calculating the peak interval T based on the peaks S of the PPG signal, and presetting the threshold R1 of the main peak misjudgment mechanism of the VPG and APG waveforms to be 0.6-0.8 times the average value of the peak interval T, and if the peak interval is less than R1, it is confirmed that there is a peak misjudgment;

[0018] The low-amplitude missed judgment mechanism is to calculate the peak interval T based on the peaks S of the PPG waveforms identified by combining a moving sliding window with an initial threshold, and preset the threshold R2 of the low-amplitude missed judgment mechanism of the VPG and APG waveforms to be 1.5-1.7 times the average value of the peak interval T, and if the current peak interval is greater than R2, it is considered that there is a low-amplitude peak missed.

[0019] Further, in step S3, the judgment mechanism of the waveform shape change classification strategy is based on extreme value calculation and curvature analysis within the signal segment to determine whether the waveform has extreme points and inflection points as the basis for classifying PPG and APG waveforms.

[0020] The two different forms of PPG waveforms are divided into two categories: with obvious double beats and without obvious double beats according to the waveform characteristics of N points and D points.

[0021] The three different forms of APG waveforms are divided into three categories: with obvious c points and d points, with relatively obvious c points and d points, and without clear c points and d points according to the waveform characteristics of c points and d points.

[0022] The judgment of whether the waveform has extreme points and inflection points is based on the identified basic reference points as the reference to divide the signal segment, first perform a first point-by-point difference operation on the signal segment, and calculate the maximum and minimum values in the segment, if both extreme points exist, the waveform is divided into a category, if not, perform a second point-by-point difference operation, calculate the curvature of the signal segment and analyze it, use the curvature to determine whether there is an inflection point, if there is, the waveform is divided into the second category, if both the extreme points and the inflection points do not exist, it is divided into the third category.

[0023] Further, in step S4, the feature point recognition method of different categories includes the concave-convex point positioning method and the slope reversal method.

[0024] The concave-convex point positioning method is to fit a straight line between two adjacent basic reference points using a mathematical formula, and calculate the point with the maximum distance from the straight line of the adjacent reference points; wherein the left side point with the maximum distance is defined as the concave point, and the right side point with the maximum distance is defined as the convex point.

[0025] The slope inversion method uses the first derivative of the signal segment (i.e., the slope of the waveform) to detect the slope sign change, and the point where the slope changes from positive to negative or from negative to positive is called the point of slope inversion, that is, the feature point to be positioned.

[0026] The present application has the advantages that the method is suitable for pathological pulse wave signals, has the ability to comprehensively identify and accurately position feature points of multiple categories of waveforms, and can more accurately capture and analyze signal characteristics in a pathological state.

[0027] On the basis of feature point identification, the present application sets up multiple judgment mechanisms that can adaptively adjust identification parameters and strategies, so that they can dynamically adjust to adapt to the physiological and pathological states of different patients according to the characteristics of PPG morphological changes caused by different cardiovascular diseases.

[0028] Other advantages, objects, and features of the present application will be set forth in part in the specification, and in part will become apparent to those skilled in the art upon examination of the following, or can be learned from practice of the present application. The objects and other advantages of the present application can be realized and attained by the methods and instrumentalities set forth in the following. BRIEF DESCRIPTION OF DRAWINGS

[0029] In order to make the objects, technical solutions, and advantages of the present application clearer, the following will describe the preferred embodiments of the present application in combination with the accompanying drawings, in which:

[0030] Figure 1 Flow chart for feature point identification and positioning of PPG and its derived signals of the present application;

[0031] Figure 2 Feature point position diagram for PPG and its derived signals of the present application;

[0032] Figure 3 Flow chart for signal preprocessing in the present application;

[0033] Figure 4 Flow chart for detection of basic reference points and mapping relationship diagram of each waveform feature point in the present application;

[0034] Figure 5 Several categories of waveform classification in the present application;

[0035] Figure 6 Feature point identification result diagram for PPG and its derived signals of the present application. DETAILED DESCRIPTION

[0036] The present application is illustrated by way of example and not limitation in the figures of the accompanying drawings, in which like references indicate similar elements, and in which: BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Wherein, the drawings are only used for example description, the representation is only schematic diagram, not real object drawing, and cannot be understood as the limitation of the present application; in order to better illustrate the embodiments of the present application, some components of the drawings are omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings can be omitted.

[0038] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it is understood that if the terms 'upper', 'lower', 'left', 'right', 'front', 'back' and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and not to indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the positional relationship described in the drawings is only used for example description, and cannot be understood as the limitation of the present application, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific situation.

[0039] Please refer to Figures 1-6 , the embodiment provides a feature point recognition and positioning method of PPG and its derivative signal, as shown in Figure 1 , the method specifically comprises the following steps:

[0040] 1) Signal preprocessing: using Butterworth low-pass filter and cubic spline interpolation method to remove high-frequency noise and baseline drift for the original PPG signal, using period segmentation method to extract the waveform in the whole period range in time domain, as shown in Figure 3 .

[0041] The cut-off frequency of the Butterworth filter is set to 10Hz based on the spectral characteristics of the pulse wave signal, which effectively eliminates the high-frequency noise component higher than 10Hz, while retaining the key low-frequency information in the signal.

[0042] The cubic spline interpolation method is to use the identified wave trough points as the reference points for interpolation, fit the whole signal, and then subtract the fitted curve from the signal filtered to remove high-frequency noise to obtain the corrected PPG signal, so as to eliminate the drift caused by the non-pulse wave component.

[0043] The period division method is to directly use the first wave trough O point and the last wave trough O point identified in the PPG waveform as the reference for waveform division.

[0044] 2) Detection of basic reference points: based on the time-domain whole-period PPG signal obtained in step 1), point-by-point difference operation is performed on the waveform data of the PPG signal, and the first derivative function VPG signal and the second derivative function APG signal of the PPG signal are calculated, and the basic reference points of the PPG, VPG and APG waveforms are detected by using the moving sliding window, threshold judgment and adaptive dynamic adjustment strategy, as shown in FIG. 2. Figure 4

[0045] The size w of the moving sliding window is set according to the formula m*fs, wherein fs is the signal sampling frequency, and m is a coefficient with a value range of 0.6-1.2. The preset amplitude threshold H is 0.6-0.8 times the maximum amplitude of the PPG signal. In this embodiment, the optimal values w=0.6*fs and H=0.7*max(PPG) are selected for identifying the wave peaks and wave troughs of the PPG, VPG and APG waveforms. The adaptive dynamic adjustment strategy includes threshold judgment of the wave peak interval and low-amplitude waveform missing detection judgment mechanism, which is specifically as follows:

[0046] In order to avoid misjudging the high-amplitude diastolic peak w and e as the main wave peak, the wave peak interval T is calculated according to the wave peak S point of the PPG signal, and the threshold R1 of the main wave peak misjudgment mechanism of the VPG and APG waveforms is preset to be 0.6-0.8 times the average value of the wave peak interval T, and the optimal value selected in this application is 0.7*mean(T). If the wave peak interval is less than R1, it is considered that there is a wave peak missing detection, and the wave peak with a smaller amplitude in the period is set to zero.

[0047] In order to avoid misjudging the high-amplitude diastolic peak w and e as the main wave peak, the wave peak interval T is calculated according to the wave peak S point of the PPG signal, and the threshold R1 of the main wave peak misjudgment mechanism of the VPG and APG waveforms is preset to be 0.6-0.8 times the average value of the wave peak interval T, and the optimal value selected in this application is 0.7*mean(T). If the wave peak interval is less than R1, it is considered that there is a wave peak missing detection, and the wave peak with a smaller amplitude in the period is set to zero.

[0048] The wave peaks and wave troughs of the PPG, VPG and APG waveforms are identified based on the above method and are respectively mapped to the zero-crossing point L of the VPG waveform, the maximum slope point E of the rising branch of the PPG waveform and the minimum slope point F of the falling branch, as shown in FIG. 3. Figure 2 ​The signal is segmented based on the peak u and the trough v points of the VPG waveform. The position of the diastolic peak w of the VPG waveform is determined by the method of sorting extreme values. The point with the maximum slope between the v point and the w point is mapped to the e point of the APG waveform. The f point is identified according to the position of the e point by the method of sorting extreme values. The e point and the f point of the APG waveform assist in positioning the N point and the D point of the PPG waveform with unobvious secondary waves.

[0049] 3) Waveform classification: The signal segment is divided based on the basic reference points identified in step 2). The classification strategy of waveform morphological changes is used for the classification of two types of PPG waveforms and three types of APG waveforms, as shown in the following table. Figure 5

[0050] The classification strategy of waveform morphological changes is based on extreme value calculation and curvature analysis in the signal segment, which is used as the basis for determining whether there is a significant extreme point and an inflection point in the waveform. Specifically:

[0051] Taking the APG waveform as an example, the basic reference points trough b and diastolic peak e of the APG waveform are used as the reference to divide the signal segment. The first point-by-point difference operation is performed on the signal segment, and the maximum value and the minimum value in the segment are calculated. If both the maximum value and the minimum value exist, the waveform is classified into a category. If they do not exist, the second point-by-point difference operation is performed, the curvature of the signal segment is calculated, and the curvature is analyzed. Whether there is an inflection point is determined by the curvature. If there is an inflection point, the waveform is classified into a second category. If neither the extreme point nor the inflection point exists, the waveform is classified into a third category.

[0052] 4) Recognition of feature points of different categories: Based on the PPG waveforms and APG waveforms of different categories obtained by the waveform classification strategy, the feature point recognition method that matches the characteristics of the waveforms is adaptively selected to recognize the N point and the D point of the PPG waveform, and the c point and the d point of the APG waveform.

[0053] The feature point recognition strategy of different categories includes concave-convex point positioning method and slope reversal method, which are as follows:

[0054] (1) The concave-convex point positioning method is to fit a straight line between the b point and the e point of the APG waveform by using a mathematical formula, and then traverse the left data points of the signal segment to calculate the point with the maximum distance to the straight line as the c point and the point with the maximum distance to the right as the d point.

[0055] (2) The slope reversal method: According to the signal from the b point to the e point of the divided segment, the first derivative (i.e. the slope of the waveform) is calculated to detect the change of the slope sign and realize the positioning of the feature points c and d. The point where the slope changes from positive to negative is the c point, and the point where the slope changes from negative to positive is the d point.

[0056] ​The feature point recognition and positioning result schematic diagram of PPG and its derivative signals based on the above steps is shown in FIG. 12. Figure 6

[0057] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the purpose and scope of the technical solutions, which should be covered in the scope of the claims of the present application.​

Claims

1. A method for feature point identification and localization of PPG and its derived signals, characterized in that, The method specifically includes the following steps: S1: Signal preprocessing: Acquire the raw PPG signal, use Butterworth low-pass filter and cubic spline interpolation to remove high-frequency noise and baseline drift, and use period segmentation method to extract waveforms within the integer period range of the time domain; S2: Detection of basic reference points: Based on the time-domain integer period PPG signal obtained in step S1, point-by-point differential operation is performed on the waveform data of the PPG signal to calculate the first derivative function VPG signal and the second derivative function APG signal of the PPG signal respectively. The basic reference points of the PPG, VPG and APG waveforms are detected by using the method of moving sliding window, threshold judgment and adaptive dynamic adjustment strategy. S3: Waveform Classification: Based on the baseline point identified in step S2, the signal segments are divided. The classification strategy based on the waveform morphology changes in the signal segments is used to effectively classify two different PPG waveforms and three different APG waveforms. S4: Feature point identification of different categories: Based on the classification strategy of waveform morphology changes obtained in step S3, different categories of PPG and APG waveforms are obtained. Feature points that match the waveform characteristics are adaptively selected to identify the N and D points of different PPG waveforms, as well as the c and d points of different APG waveforms. Among them, the N point and D point are the bisection notch point and the diastolic peak point of the PPG waveform, respectively; the c point is the late contraction re-increase wave of the APG waveform, and the d point is the late contraction re-attenuation wave of the APG waveform.

2. The method for feature point recognition and localization according to claim 1, characterized in that, In step S2, the basic reference points include the peak S and trough O of the PPG waveform, the peak u and trough v of the VPG waveform, the peak a and trough b of the APG waveform, and the characteristic points of each waveform's peaks and troughs mapped to other waveforms, namely the maximum slope point E of the rising branch and the minimum slope point F of the falling branch of the PPG waveform, the zero-crossing point L of the VPG waveform, the diastolic peak point w of the VPG waveform, and the additional peak point e of the late contraction period of the APG waveform.

3. The method for feature point recognition and localization according to claim 2, characterized in that, In step S2, the width value W of the sliding window is set according to the following formula: W = m * fs Where fs is the signal sampling frequency, m is a coefficient with a value range of 0.6 to 1.2; the preset amplitude threshold H is 0.6 to 0.8 times the largest amplitude in the waveform data, and identifies the peaks and troughs of PPG, VPG and APG waveforms; The adaptive dynamic adjustment strategy includes threshold judgment for the interpeak period and a low-amplitude missed detection judgment mechanism.

4. The method for feature point recognition and localization according to claim 3, characterized in that, In step S2, the threshold judgment of the inter-peak period is to use a combination of a moving sliding window and an initial threshold to identify the peaks and troughs of PPG, VPG and APG signals. Then, the inter-peak period T is calculated based on the peak S point of the PPG signal. The threshold R1 of the main wave peak misjudgment mechanism of VPG and APG waveforms is preset to 0.6 to 0.8 times the average value of the inter-peak period T. If the inter-peak period is less than R1, it is confirmed that there is a false peak detection. The low-amplitude missed detection judgment mechanism is based on the peak S point of the PPG waveform identified by a combination of a moving sliding window and an initial threshold, and the peak interval T is calculated. The threshold R2 of the low-amplitude missed detection judgment mechanism for VPG and APG waveforms is preset to 1.5 to 1.7 times the average value of the peak interval T. If the current peak interval is greater than R2, it is considered that there is a low-amplitude peak missed in the middle.

5. The method for feature point recognition and localization according to claim 1, characterized in that, In step S3, the judgment mechanism of the waveform morphology change classification strategy is based on extreme value calculation and curvature analysis within the signal segment to determine whether there are extreme points and inflection points in the waveform, which serves as the basis for PPG and APG waveform classification. The two different PPG waveforms are classified into two categories based on the waveform characteristics at the N and D points: those with obvious diphtheria waves and those without obvious diphtheria waves. The three different forms of APG waveforms are classified into three categories based on the waveform characteristics of points c and d: those with significant points c and d, those with relatively obvious points c and d, and those without clear points c and d. The determination of whether a waveform has extreme points and inflection points is based on dividing the signal segment according to the identified baseline point. First, a point-by-point difference operation is performed on the signal segment, and the maximum and minimum values ​​in the segment are calculated. If both extreme points exist, the waveform is classified into one category; if not, a second point-by-point difference operation is performed to calculate and analyze the curvature of the signal segment. The curvature is used to determine whether an inflection point exists. If it exists, the waveform is classified into a second category; if neither extreme points nor inflection points exist, it is classified into a third category.

6. The method for feature point recognition and localization according to claim 1, characterized in that, In step S4, the different types of feature point recognition methods include the concave-convex point localization method and the slope inversion method; The concave-convex point positioning method uses mathematical formulas to fit a straight line between two adjacent basic reference points and calculates the point with the largest distance from the signal of the adjacent reference points to the straight line; wherein, the point with the largest distance on the left is defined as a concave point, and the point with the largest distance on the right is defined as a convex point. The slope reversal method uses the first derivative of the signal segment, i.e. the slope of the waveform, to detect changes in the sign of the slope. The point where the slope changes from positive to negative or from negative to positive is called the slope reversal point, which is also the feature point that needs to be located.

Images